Introduction

Despite there is a wide interest in climate change mitigation (Andre et al. 2024; Falkner 2016; Mayer, Shames, and Wronski 2017), global coverage surveys only began in recent years; e.g. WRP (2019), PCV (2021).

• Generally, survey data about climate change is:
  • Skewed: Few low-income countries, small islands.
  • Inconsistent: Variations in wording, response scales, and survey structure.
  • Limited: Longitudinal data are rare (Eurobarometer).

Introduction

We aim to harmonize disparate survey data on climate attitudes into a single, comparable latent measure. With this is mind, we will generate three complementary datasets:

1. Cross-sectional: Maximize country coverage.
   
2. Country-year data: Maximize matrix consistency.
   
3. Longitudinal: Maximize temporal depth.

Data collection

• Questions about climate change attitudes from 12 major global/regional surveys:
  • G: World Risk Poll, WVS, ISSP, etc.
  • R: Eurobarometer, Afrobarometer, CAB, etc.
• 8,000+ item-country-year observations (2010-2024)
  • Over 3.9 million of individual responses.
  • 173 countries

Data collection

Figure 1: Global coverage of climate change attitude survey data (2010–2024)

Exploratory analysis

Items can represent different dimensions of CC support.

• PCA between high coverage items reveals unidimensional structure of the data.

Figure 3: Parallel analysis scree plot for PCA (27 items; N > 40)

Exploratory analysis

Item selection criteria:

• Retained medium- to high-coverage items with a correlation ≥ 0.3 to the principal component.
  • Ensures conceptual coherence with the latent dimension of climate support
    
• Included low-coverage items (fewer than 15 countries) regardless of correlation to preserve regional diversity.
  • Central Asia Barometer, Arab Barometer, etc.

Statistical model

Bayesian ordinal Item Response Theory (IRT) with partial pooling (e.g. Claassen 2019).

• Handles ordinal data.
• Useful with missing data and uneven coverage.
• Temporal smoothing to track trends from 2010–2024

Cross-sectional dataset (2019-24)

167 countries; world GDP (99.5%); population (98.6%).

Reasonably balanced: 60%-60% dem, 24k-18k GDPcap.

Country-year panel (2010–2024)

Nearly 72 countries (57% global GDP, 28% pop).

Unbalanced: 70%-40% democracies, 31k-13k GDPc

Extended time series (1990s–2024)

Nearly 50 countries.

Unbalanced: Primarily OECD, democracies and high GDPc.

Conclusion (I)

This research presents a harmonized and globally representative dataset of public attitudes toward climate change.

• Addresses major limitations in existing data:
  • Skewed coverage (overrepresenting high-income democracies).
  • Inconsistent survey formats and item wording.
  • Sparse and fragmented longitudinal data.

Conclusion (II)

• Applies a Bayesian ordinal IRT model to estimate a latent measure of climate support across countries and years.

• Outputs three complementary datasets.

  • Cross-sectional: over 165 countries.
  • Country-year panel: consistent data 2010-2024.
  • Longitudinal series: historical trends for ~50 countries, mainly in the OECD